Can you fully explain how SCSI target/LUNs are mapped to Fibre Channel LUNs with persistent binding (across boots) on Fibre Channel switched networks, SAN for AIX, SUN, and NT systems? I believe it is related to the HBA adapter, but am uncertain how to ensure guaranteed persistent associations.
Could you also indicate if there are any special considerations for EMC Symmetrix storage with EMC Volume Logix software?
I assume the question is related to what happens if a SAN goes lights-out (disaster occurs) and has to be rebuilt. For starters, there is no persistent "bind" between end nodes - systems and storage as the case is. Fibre Channel is a network that communicates using serial SCSI as an application level protocol. End nodes can generally access any other end node unless they are blocked from doing so. Applications such as file systems and databases may be able to be configured to access certain storage nodes, but this is not a Fibre Channel network function. Fibre Channel SANs are commonly implemented where connections between end nodes can be segregated using "zoning" in switches or "masking" (filtering) to restrict access from certain other potentially dangerous nodes.
Persistence in Fibre Channel SANs is provided by using 64-bit worldwide names-set at the factory on each network entity. Switches keep name service tables that identify all entities uniquely. When a node reconnects to the fabric, it logs into the fabric and the switch is able to identify it by its worldwide name. I believe most switches keep these name service tables in non-volatile memory, but I do not know if there is any way to export them or back them up in case they might need to be restored in a replacement switch. If a bunch of new HBAs are introduced following a disaster, these name-service tables would have to be rebuilt. In that case, switch zones may need to be reconfigured in order to ensure new HBAs do not access the wrong storage.
FC loops have a 5-step process for establishing persistence following a loop initialization. This depends on all the entities in the loop being good network citizens.
It's easy to address the EMC part of the question because I don't know this software or what it does. That said, I believe EMC probably has a fairly bullet-proof mechanism for recreating virtual LUNs and target IDs following some sort of crash. I do not know the Symmetrix architecture, but I believe it has non-volatile storage that maintains its own configuration information. In fact, I believe it is also possible that EMC corporation can keep this data as a service for its customers and can be tied into "call home" services.
I hope this answer helps out.
This was first published in April 2001